I saw in the code that we just log an error if the connection to headscale cannot be established. But then we could have metal-api replicas after an rolling update which have no headscale client initialized and these will consistently panic on all headscale endpoints.
Found through GKE log console that this can happen:
I saw in the code that we just log an error if the connection to headscale cannot be established. But then we could have metal-api replicas after an rolling update which have no headscale client initialized and these will consistently panic on all headscale endpoints.